CN118799398A - Visual localization method based on alignment of 3D LoD map and neural wireframe - Google Patents

Abstract

The present application relates to a visual positioning method based on three-dimensional LoD map and neural wireframe alignment. The method comprises: uniformly sampling four degrees of freedom with the initial posture as the center, generating posture hypotheses in four directions respectively; calculating the straight line alignment cost of the posture hypothesis and the pre-constructed three-dimensional wireframe points according to the neural wireframe alignment method, combining the straight line alignment costs in four directions in a grid manner to calculate the posture cost volume and then calculating the probability distribution volume; determining the posture sampling range of the next layer according to the probability distribution volume to generate the posture hypothesis of the next layer, and taking the selected posture obtained at the last feature level as the candidate selected posture; mapping the multi-level features, using the mapped features and the three-dimensional wireframe points to design the optimization objective function of the candidate selected posture and then solving it to obtain the final posture. The use of this method can improve the visual positioning accuracy and low memory.

Description

Visual localization method based on alignment of 3D LoD map and neural wireframe

Technical Field

The present application relates to the field of visual positioning technology, and in particular to a visual positioning method based on alignment of a three-dimensional LoD map with a neural wireframe.

Background Art

Visual localization is the process of determining the position and orientation, i.e., the camera pose, of a given image relative to a known map. It is a fundamental problem in many 3D computer vision applications, ranging from autonomous driving and navigation of unmanned aerial vehicles (UAVs) to augmented reality. Current state-of-the-art visual localization methods typically involve matching pixels in a query image with points in a pre-built high-quality 3D map, which is usually derived from structure from motion (SfM) or a 3D textured model. Subsequently, the camera pose is typically computed using a perspective n-point RANSAC technique.

However, building high-quality 3D maps using photogrammetry is expensive worldwide and requires frequent data updates to capture temporal changes in visual appearance. In addition, the storage cost of these 3D maps is high, which poses a major challenge to terminal deployment on various mobile devices such as mobile phones and drones. In addition, high-resolution 3D maps disclose detailed information about the location area, raising key issues regarding homeland security and privacy protection.

Summary of the invention

Based on this, it is necessary to provide a visual positioning method based on alignment of three-dimensional LoD map and neural wireframe, which can improve the visual positioning accuracy and achieve low memory storage, in order to address the above technical problems.

A visual positioning method based on alignment of a three-dimensional LoD map with a neural wireframe, the method comprising:

Obtain the query image taken by the drone on the 3D city LoD map and the prior attitude of the drone sensor; build a visual positioning model; the visual positioning model includes a feature extraction module, an attitude selection module and an attitude optimization module;

In the feature extraction module, the query image is subjected to feature extraction at multiple levels according to the convolutional neural network to obtain multi-level features;

In the posture selection module, the initial posture on the current feature level is defined, and the four degrees of freedom are uniformly sampled with the initial posture as the center, and posture hypotheses are generated in four directions respectively; the straight line alignment cost of the posture hypothesis and the pre-constructed three-dimensional wireframe points is calculated according to the neural wireframe alignment method, and the straight line alignment costs in four directions are combined in a grid manner to obtain the posture cost volume; the posture cost volume is calculated according to the softmax function to obtain the probability distribution volume; the argmax operation is performed on the probability distribution volume to obtain the selected posture; the variance of the probability distribution volume of the current feature level at the current level is used to determine the posture sampling range of the next layer to generate the posture hypothesis of the next layer, and the selected posture obtained at the last feature level is used as the candidate selected posture; the initial posture on the first feature level is the prior posture;

In the posture optimization module, the multi-level features are mapped, and the optimization objective function of the candidate selection posture is designed using the mapped features and 3D wireframe points. The optimization objective function is solved according to the Gauss-Newton method to obtain the final posture;

The posture selection module is trained according to a preset posture selection loss function to obtain a trained posture selection module; the posture optimization module is optimized using a preset posture optimization loss function to obtain a trained posture optimization module;

Use the trained visual localization model to perform visual localization on the input image.

In one embodiment, defining an initial pose at a current feature level includes:

The initial pose is defined at the current feature level as where (x _l ,y _l ,z _l ) represents the translation in three-dimensional space, They represent the yaw angle, pitch angle and roll angle respectively, and l represents the feature level number.

In one embodiment, the four degrees of freedom are uniformly sampled with the initial posture as the center, and posture hypotheses are generated in four directions, including:

The four degrees of freedom are uniformly sampled with the initial posture as the center, and the postures generated in four directions are assumed to be

Among them, d∈((x, y, z, θ), (x _l , y _l , z _l ) represents the translation in three-dimensional space, θ _l represents the yaw angle, r _l represents the sampling range, m _l represents the number of samples, and l represents the feature level number.

In one embodiment, calculating a straight line alignment cost of a pose hypothesis and pre-constructed three-dimensional wireframe points according to a neural wireframe alignment method includes:

According to the neural wireframe alignment method, the straight line alignment cost of the pose hypothesis and the pre-constructed 3D wireframe points is calculated as

in, represents the pose hypothesis, F _l represents the features of layer l, and _Pi represents the 3D wireframe points.

In one embodiment, the calculation process of the variance of the probability distribution volume of the current feature level at the current level includes:

Due to the posture assumption The pose cost volume C _l and the probability distribution volume P _l have the same data structure. They are flattened and indexed by t. The variance v _l at layer l is calculated as

Where l represents the current feature level number, P _l-1 represents the probability distribution volume of the l-1 layer, Represents the selected pose of layer l-1.

In one embodiment, the variance of the probability distribution volume of the current feature level at the current level is used to determine the gesture sampling range of the next level, including:

The standard deviation is calculated based on the variance: The standard deviation is used to determine the gesture sampling range as r _l =2λ·σ _l , where λ is a hyperparameter for adjusting the length of the sampling range, and l represents the feature level number.

In one embodiment, the optimized objective function of the candidate selection pose is designed using the mapped features and the three-dimensional wireframe points, including:

The optimization objective function of designing candidate selection poses using the mapped features and 3D wireframe points is:

Among them, F _rf represents the mapped feature, _Pi represents the three-dimensional wireframe point, ξ ^* = (R ^* , t ^* ) represents the final posture, and Π is the projection operation.

In one embodiment, the preset posture selection loss function is

Where l∈{1,2,3}, p _l represents the probability distribution volume, represents the ground truth pose.

In one embodiment, the preset posture optimization loss function is

Among them, the ground truth posture ρ represents the Huber robust kernel, ξ ^* = (R ^* , t ^* ) represents the final pose, and Π is the projection operation.

The above-mentioned visual positioning method based on alignment of three-dimensional LoD map with neural wireframe, this application realizes visual positioning by constructing a visual positioning model on the three-dimensional city LoD map, and taking the query image taken by the drone and its sensor prior pose as input. Compared with the existing complex 3D representation, this application uses the detail level LoD map to estimate the pose of the drone, providing a simple, accessible and privacy-friendly scene representation. In the constructed visual positioning model, multi-layer feature extraction is first performed on the query image, and a hierarchical pose estimation scheme is proposed, which uses multiple small pose volumes instead of large cost volumes to gradually calculate high-quality poses in a coarse-to-fine manner. In the hierarchical process, an adaptive sampling strategy is adopted, in which the uncertainty based on variance in the previous stage affects the sampling range of the next stage to construct the pose cost volume. This adaptive process induces a reasonable and fine-grained pose space division, which significantly improves the final pose output, improves the visual positioning accuracy and achieves low memory storage. In addition, after the coarse-to-fine attitude estimation stage, the multi-level features are mapped, and the optimized objective function of the candidate selection attitude is designed using the mapped features and 3D wireframe points. The optimized objective function is solved according to the Gauss-Newton method to obtain the final attitude, which corrects the slight error in the initial gravity prior and improves the overall attitude accuracy.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic diagram of a flow chart of a visual positioning method based on alignment of a three-dimensional LoD map with a neural wireframe in one embodiment;

FIG2 is a schematic diagram of a framework of a visual positioning method in one embodiment;

FIG. 3 is a schematic diagram of an uncertainty sampling range estimation process in one embodiment.

DETAILED DESCRIPTION

In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

In one embodiment, as shown in FIG. 1 and FIG. 2 , a visual positioning method based on alignment of a three-dimensional LoD map with a neural wireframe is provided, comprising the following steps:

Step 102, obtaining a query image taken by the drone on the three-dimensional city LoD map and a priori posture of the drone sensor; constructing a visual positioning model; the visual positioning model includes a feature extraction module, a posture selection module and a posture optimization module.

Step 104, in the feature extraction module, feature extraction is performed on the query image at multiple levels according to the convolutional neural network to obtain multi-level features.

A convolutional neural network with UNet is used to extract multi-level features from the query image, maintaining a high-dimensional feature map to encapsulate the rich visual information of each layer, abstracting the dimension of the feature map and reducing it to 1, where each pixel in the map represents the possibility of becoming a wireframe. The obtained multi-level features are used Represents, where l = {1, 2, 3} is the level index.

Step 106, in the posture selection module, define the initial posture on the current feature level, uniformly sample the four degrees of freedom with the initial posture as the center, and generate posture hypotheses in four directions respectively; calculate the straight line alignment cost of the posture hypothesis and the pre-constructed three-dimensional wireframe points according to the neural wireframe alignment method, and combine the straight line alignment costs in four directions in a grid manner to obtain the posture cost volume; calculate the posture cost volume according to the softmax function to obtain the probability distribution volume; perform argmax operation on the probability distribution volume to obtain the selected posture; use the variance of the probability distribution volume of the current feature level at the current level to determine the posture sampling range of the next layer to generate the posture hypothesis of the next layer, and use the selected posture obtained at the last feature level as the candidate selected posture; the initial posture on the first feature level is the prior posture.

After feature extraction, a cost volume is built for various pose hypotheses sampled around the pose, and then the pose with the highest probability is selected at each level. To ensure effective sampling, the uncertainty of the pose selection at the current layer is used to determine the range of pose sampling at the next layer. First, the initial pose on the current feature level is defined. Considering that the pitch and roll of the gravity direction of the previous data have high accuracy, only four degrees of freedom are uniformly sampled around the initial pose, and pose hypotheses are generated in four directions respectively. The pose hypotheses are generated by sampling four degrees of freedom (including position and yaw angle) around the prior, assuming that the gravity direction provided by the inertial unit has a small error. Based on the generated pose hypotheses, the LoD building wireframe is projected onto the query image plane. Each pose hypothesis is then scored by the alignment between the predicted wireframe and the predicted wireframe. By combining these alignment costs in a grid manner on four different dimensions (x, y, Z, θ), a 4D pose cost volume C _l with dimensions [m _l (x)×m _l (y)×m _l (z)×m _l (θ)] is obtained. Perform the softmax function on C _l to obtain the probability distribution volume Pl, derive the probability density of the posture, which can be used for posture estimation through classification (similar to classification classification) or regression (similar to position expectation) operations, and perform argmax operation on P _l to select the posture with the maximum probability The whole process interprets the output of pose estimation as a probability distribution parameterized by a learnable network to predict a 2D wireframe of the query image. In the uncertain sampling range estimation process, the pose selection uncertainty of the previous layer is used to determine the sampling range of the current layer in the coarse-to-fine process. The sampling range of the current level. This strategy allows the pose sampling space to be gradually subdivided, thereby improving the accuracy of pose selection. More specifically, for l = 1, the pose sampling range is defined by carefully examining the maximum divergence between the prior and GT poses in the UAVD4L-LoD dataset. The sampling range of (x, y, z, θ) is established as [r ₁ (x), r ₁ (y), r ₁ (z), r ₁ (z)] (= ([r _p (x), r _p (y), r _p (z), r _p (z)]. For l = {2,3}, the variance of the probability distribution volume P _l-1 at l-1 is used to determine the pose sampling range r _l , where, due to the pose assumption The pose cost volume C _l and the probability distribution volume P _l have the same data structure, which are flattened and indexed by t. The variance v _l at the lth layer is calculated as

symbol represents the subtraction operation applied to the (x, y, z, θ) directions respectively. The corresponding standard deviation is calculated as The calculated pose sampling range is r _l =2λ·σ _l , where λ is a hyperparameter that adjusts the length of the sampling range. The visualization of this uncertain sampling range estimation process is shown in Figure 3.

Through the hierarchical pose estimation scheme of this application, multiple small pose volumes are used instead of large cost volumes to gradually calculate high-quality poses in a coarse-to-fine manner. In the hierarchical process, an adaptive sampling strategy is adopted, in which the variance-based uncertainty of the previous stage affects the sampling range of the next stage to construct the pose cost volume. This adaptive process induces a reasonable and fine-grained partitioning of the pose space, which significantly improves the final pose output, improves visual positioning accuracy and achieves low memory storage.

Step 108, in the posture optimization module, the multi-level features are mapped, and the optimization objective function of the candidate selection posture is designed using the mapped features and the three-dimensional wireframe points. The optimization objective function is solved according to the Gauss-Newton method to obtain the final posture.

Based on the posture selected in the previous stage The refined wireframe probability map _F3 further extracted from the feature map _Frf by the post-processing convolutional network is used to optimize the pose ξ ^* = (R ^* , t ^* ) to align the 3D wireframe with the 2D predicted wireframe. The optimization objective function of the candidate selected pose is designed using the mapped features and 3D wireframe points:

Where _Pi is a 3D wireframe point and Π is a projection operation. Minimizing this function allows the projected 3D wireframe point to move to the 2D position with a higher prediction probability. The pose update formula of ξ ^* derived from the Gauss-Newton method is

R ^* ＝R ^* ·exp(△ _ξr )

In the formula, is the six-dimensional transformation vector, is the rotation component, is the translation component. Using Lie algebra The exponential mapping of , transforms the rotation component △ξ _r into a 3×3 rotation matrix. Where _Ji represents the Jacobian matrix of the residual function _fi with respect to the pose parameters.

Step 110, training the posture selection module according to a preset posture selection loss function to obtain a trained posture selection module; optimizing the posture optimization module using a preset posture optimization loss function to obtain a trained posture optimization module; and performing visual positioning on the input image using the trained visual positioning model.

The pre-set posture selection loss function is used to constrain the negative log-likelihood loss of the posture selection module, which means that the closer the selected posture is to the true value, the greater the probability corresponding to the posture, which can improve the accuracy of posture selection; the pre-set posture optimization loss function is the loss function of the posture optimization module (Newton iterative optimization), which uses the reprojection error to constrain the adjustment of the posture. By setting the combination of the above two loss functions, the final posture result can be optimized from coarse to fine.

In the above-mentioned visual positioning method based on alignment of three-dimensional LoD map with neural wireframe, this application realizes visual positioning by constructing a visual positioning model on the three-dimensional city LoD map, and taking the query image taken by the drone and its sensor prior pose as input. Compared with the existing complex 3D representation, this application uses the detail level LoD map to estimate the pose of the drone, providing a simple, accessible and privacy-friendly scene representation. In the constructed visual positioning model, multi-layer feature extraction is first performed on the query image, and a hierarchical pose estimation scheme is proposed, which uses multiple small pose volumes instead of large cost volumes to gradually calculate high-quality poses in a coarse-to-fine manner. In the hierarchical process, an adaptive sampling strategy is adopted, in which the uncertainty based on variance in the previous stage affects the sampling range of the next stage to construct the pose cost volume. This adaptive process induces a reasonable and fine-grained pose space division, which significantly improves the final pose output, improves the visual positioning accuracy and achieves low memory storage. In addition, after the coarse-to-fine attitude estimation stage, the multi-level features are mapped, and the optimized objective function of the candidate selection attitude is designed using the mapped features and 3D wireframe points. The optimized objective function is solved according to the Gauss-Newton method to obtain the final attitude, which corrects the slight error in the initial gravity prior and improves the overall attitude accuracy.

In one embodiment, defining an initial pose at a current feature level includes:

The initial pose is defined at the current feature level as where (x _l ,y _l ,z _l ) represents the translation in three-dimensional space, They represent the yaw angle, pitch angle and roll angle respectively, and l represents the feature level number.

In one embodiment, the four degrees of freedom are uniformly sampled with the initial posture as the center, and posture hypotheses are generated in four directions, including:

The four degrees of freedom are uniformly sampled with the initial posture as the center, and the postures generated in four directions are assumed to be

Among them, d∈((x, y, z, θ), (x _l , y _l , z _l ) represents the translation in three-dimensional space, θ _l represents the yaw angle, r _l represents the sampling range, m _l represents the number of samples, and l represents the feature level number.

In one embodiment, calculating a straight line alignment cost of a pose hypothesis and pre-constructed three-dimensional wireframe points according to a neural wireframe alignment method includes:

According to the neural wireframe alignment method, the straight line alignment cost of the pose hypothesis and the pre-constructed 3D wireframe points is calculated as

in, represents the pose hypothesis, F _l represents the features of layer l, _Pi represents the 3D wireframe points, and [·] is the sub-pixel interpolation search.

In a specific embodiment, the construction process of three-dimensional wireframe points is prior art and will not be described in detail in this application.

In one embodiment, the calculation process of the variance of the probability distribution volume of the current feature level at the current level includes:

Due to the posture assumption The pose cost volume C _l and the probability distribution volume P _l have the same data structure. They are flattened and indexed by t. The variance v _l at layer l is calculated as

Where l represents the current feature level number, P _l-1 represents the probability distribution volume of the l-1 layer, Represents the selected pose of layer l-1.

In one embodiment, the variance of the probability distribution volume of the current feature level at the current level is used to determine the gesture sampling range of the next level, including:

The standard deviation is calculated based on the variance: The standard deviation is used to determine the gesture sampling range as r _l =2λ·σ _l , where λ is a hyperparameter for adjusting the length of the sampling range, and l represents the feature level number.

In one embodiment, the optimized objective function of the candidate selection pose is designed using the mapped features and the three-dimensional wireframe points, including:

The optimization objective function of designing candidate selection poses using the mapped features and 3D wireframe points is:

Among them, F _rf represents the mapped feature, _Pi represents the three-dimensional wireframe point, ξ ^* = (R ^* , t ^* ) represents the final posture, and Π is the projection operation.

In one embodiment, the preset posture selection loss function is

Where l∈{1,2,3}, p _l represents the probability distribution volume, represents the ground truth pose.

In one embodiment, the preset posture optimization loss function is

Among them, the ground truth posture ρ represents the Huber robust kernel, ξ ^* = (R ^* , t ^* ) represents the final pose, and Π is the projection operation.

In a specific embodiment, during the model training process, a random seed is set to retain the 3D wireframe points {P _i } to 2000 points, and due to CUDA memory-related constraints, for level(l(＝1,2,3, the pose sampling numbers m _l (x), m _l (y), m _l (z), m _l (θ) are uniformly assigned to [13,7,3]. The image size of the UAVD4L dataset is (512,480), and the image size of the Swiss-EPFL dataset is (720,480). The pose sampling range of level(1 is set to [10,10,30,7.5], which refers to [r _p (x), r _p (y), r _p (z), r _p (z)]. The hyperparameter λ is fixed to 1.5. For the UAVD4L-(lod dataset, a subset of synthetic images in UAVD4L[64] is used as training data, which includes buildings. For Swiss-EPFL, the model is trained by combining synthetic images LHS from the CrossLoc project and real query images. During inference, the following changes are made to change the retrieved discrete points in the 3D wireframe at intervals of 1 meter. The number of pose samples is increased to [m _l (x), m _l (y), m _l (z), m _l (θ)](=[10,10,30,8]. λ is set to 0.8. The training and inference of the entire network are performed using 2 NVIDIA(RTX(4090(GPUs).

It should be understood that, although the various steps in the flowchart of FIG. 1 are shown in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in FIG. 1 may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in the present application can include non-volatile and/or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) (DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims (9)

1. A visual localization method based on alignment of a three-dimensional LoD map with a neural wireframe, the method comprising:

acquiring a query image shot by the unmanned aerial vehicle on a three-dimensional city LoD map and a priori posture of a sensor of the unmanned aerial vehicle; constructing a visual positioning model; the visual positioning model comprises a feature extraction module, a gesture selection module and a gesture optimization module;

The feature extraction module is used for extracting features of the query image on a plurality of layers according to a convolutional neural network to obtain multi-level features;

Defining an initial gesture on the current feature level at the gesture selection module, uniformly sampling four degrees of freedom by taking the initial gesture as a center, and respectively generating gesture assumptions in four directions; calculating the straight line alignment cost of the gesture hypothesis and the pre-constructed three-dimensional line frame point according to a nerve line frame alignment method, and combining the straight line alignment cost in four directions in a grid mode to obtain a gesture cost volume; calculating the gesture cost volume according to a softmax function to obtain a probability distribution volume; performing argmax operation on the probability distribution volume to obtain a selected pose; determining a gesture sampling range of a next layer by utilizing the variance of the probability distribution volume of the current feature level at the current level to generate a next layer of gesture hypothesis, and taking the selected gesture obtained by the last feature level as a candidate selected gesture; the initial pose on the first feature level is an a priori pose;

Mapping the multi-level features in the gesture optimization module, designing a candidate gesture optimization objective function by using the mapped features and the three-dimensional wire frame points, and solving the optimization objective function according to a Gauss Newton method to obtain a final gesture;

training the gesture selection module according to a preset gesture selection loss function to obtain a trained gesture selection module; optimizing the attitude optimization module by using a preset attitude optimization loss function to obtain a trained attitude optimization module;

And performing visual positioning on the input image by using the trained visual positioning model.

2. The method of claim 1, wherein defining an initial pose at a current feature level comprises:

defining an initial pose as on a current feature level Where (x _l,y_l,z_l) represents translation in three-dimensional space,Respectively representing yaw angle, pitch angle and roll angle, and l represents a characteristic hierarchy number.

3. The method of claim 1, wherein uniformly sampling four degrees of freedom centered on the initial pose, generating pose hypotheses in four directions, respectively, comprises:

uniformly sampling four degrees of freedom by taking the initial gesture as the center, and generating gesture assumptions in four directions respectively as

Where d ε (x, y, z, θ), (x _l,y_l,z_l) represents translation in three-dimensional space, θ _l represents yaw angle, r _l represents sampling range, m _l represents the number of samples, and l represents feature level number.

4. The method of claim 1, wherein calculating a straight line alignment cost of the pose hypothesis and a pre-constructed three-dimensional wireframe point according to a neural wireframe alignment method comprises:

calculating the straight line alignment cost of the posture assumption and the pre-constructed three-dimensional wire frame point according to the nerve wire frame alignment method

Wherein, Representing the pose hypothesis, F _l represents the features of the l-layer, and P _i represents the three-dimensional wireframe points.

5. The method of claim 1, wherein the calculating of the variance of the probability distribution volume of the current feature level at the current level comprises:

due to the pose assumption The pose cost volume C _l and the probability distribution volume P _l have the same data structure, they are flattened and indexed by t, and the variance v _l at the first layer is calculated as

Where l represents the current feature level number, P _l-1 represents the probability distribution volume of the l-1 layer,Representing the selected pose of the l-1 layer.

6. The method of claim 1, wherein determining the pose sampling range for the next layer using the variance of the probability distribution volume of the current feature level at the current level comprises:

calculating standard deviation from the variance as And determining the attitude sampling range as r _l＝2λ·σ_l by using the standard deviation, wherein lambda is a super parameter for adjusting the length of the sampling range, and l represents the characteristic hierarchy number.

7. The method of claim 1, wherein designing an optimization objective function for candidate selection poses using the mapped features and the three-dimensional wireframe points, comprises:

The optimization objective function of the three-dimensional wire frame point design candidate selection gesture is selected by using the mapped characteristics and the three-dimensional wire frame point design candidate selection gesture as follows

Wherein F _rf denotes the mapped feature, P _i denotes the three-dimensional wireframe point, ζ ^*＝(R^*,t^*) denotes the final pose, and pi is the projection operation.

8. The method according to claim 1, wherein the pre-set pose selection loss function is

Where l ε {1,2,3}, p _l represents the probability distribution volume,Representing the ground real attitude.

9. The method according to claim 1, wherein the pre-set pose optimization loss function is

Wherein, the ground is in a true postureΡ represents the Huber robust kernel, ζ ^*＝(R^*,t^*) represents the final pose, and ζ is the projection operation.